[R] Fwd: Potential Issue with lm.influence

Eric Bridgeford er|cwb95 @end|ng |rom gm@||@com
Tue Apr 2 11:12:11 CEST 2019


Hi R core team,

I experienced the following issue with the attached data/code snippet,
where the studentized residual for a single observation appears to be NaN
given finite predictors/responses, which appears to be driven by the
glm.influence method in the stats package. I am curious to whether this is
a consequence of the specific implementation used for computing the
influence, which it would appear is the driving force for the NaN influence
for the point, that I was ultimately able to trace back through the
lm.influence method to this specific line
<https://github.com/SurajGupta/r-source/blob/a28e609e72ed7c47f6ddfbb86c85279a0750f0b7/src/library/stats/R/lm.influence.R#L67>
which
calls C code which calls iminfl.f
<https://github.com/SurajGupta/r-source/blob/master/src/library/stats/src/lminfl.f>
(I
don't know fortran so I can't debug further). My understanding is that the
specific issue would have to do with the leave-one-out variance estimate
associated with this particular point, which it seems based on my
understanding should be finite given finite predictors/responses. Let me
know. Thanks!

Sincerely,

-- 
Eric Bridgeford
ericwb.me


More information about the R-help mailing list